IAAR-Shanghai/phase_tree_data

PHASE-Tree数据集是一个基于心理学的层次化属性结构化演化树数据集，包含8个角色对话数据集，覆盖短期和长期两种评估设置。数据集提供原始语料库和经过完整处理的评估数据，用于文本生成和对话建模任务，特别关注角色扮演、角色模仿和人物设定。数据集支持多种配置文件表示方法（从M1到M6），包括静态和动态演化树结构，适用于训练和评估对话系统。数据来源于多个公开数据集，如RAIDEN、CharacterEval、SimsConv等，并分为训练、随机测试和分布外测试等分割。PHASE-Tree Data is a dataset for Psychology-grounded Hierarchical Attribute-Structured Evolving Tree, covering 8 character-dialogue datasets across short-term and long-term evaluation settings. It includes raw source corpora and fully processed evaluation data for text-generation and dialogue-modeling tasks, with a focus on role-play, character-impersonation, and persona. The dataset supports multiple profile representation methods (M1 to M6), including static and dynamic evolving tree structures, and is designed for training and evaluating dialogue systems. Data is sourced from public datasets such as RAIDEN, CharacterEval, SimsConv, etc., and is split into train, random test, and out-of-distribution test sets.